Efficient regression of general-activity human poses from depth images

By Ross B. Girshick, Jamie Shotton, Pushmeet Kohli, Antonio Criminisi, Andrew William Fitzgibbon. 2011

434 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Space-Time Representation of People Based on 3D Skeletal Data: A Review (2016, methodology): “An extended work was published in [51], with both accuracy and speed improved.”
Flowing ConvNets for Human Pose Estimation in Videos (2015, neutral): “More recent work has tackled pose estimation holistically: initially with Random Forests on depth data [12, 29, 31, 34] and RGB [3, 4, 24], and most recently with convolutional neural networks.”
Structured Prediction of 3D Human Pose with Deep Neural Networks (2016, background): “3D human pose can now be estimated reliably by training algorithms to exploit depth data [7, 27] or video sequences [3, 11, 30].”
Free-Viewpoint Video of Human Actors Using Multiple Handheld Kinects (2013, methodology): “Recently , this approach has been augmented with a regression method to enable the algorithm to localize joints also under occlusions [43].”
Deep Learning for Image Super-Resolution: A Survey (2019, background): “Depth maps record the distance between the viewpoint and the objects in the scene, and the depth information plays important roles in many tasks such as pose estimation [150], [151], [152], semantic segmentation [153], [154], etc.”
Exploring the Trade-off Between Accuracy and Observational Latency in Action Recognition (2012, background): “The recent availability of commodity RGB-D sensors, such as the Microsoft Kinect, has led to increased research in the application of human pose data (Girshick et al. 2011; Shotton et al. 2011).”
Fusing 2D Uncertainty and 3D Cues for Monocular Body Pose Estimation (2016, neutral): “Discriminative methods aim at predicting 3D pose directly from the input data, may it be single images [26, 27, 34, 35, 36, 46, 49, 58, 67], depth images [22, 44, 52], or short image sequences [59].”
Decision Forests: A Unified Framework for Classification, Regression, Density Estimation, Manifold Learning and Semi-Supervised Learning (2012, result): “Decision forests compare favorably with respect to other techniques [15] and have lead to one of the biggest success stories of computer vision in recent years: the Microsoft Kinect for XBox 360 [39, 91, 66].”
Pose for Action - Action for Pose (2016, background): “Random regression forests have been proven to be robust for the task of human pose estimation [5], [24], [27].”
VNect (2017, background): “sometimes better than, state-of-the-art single RGB-D methods [Girshick et al. 2011], even commercial ones [Microsoft Corporation 2015].”
Structured Prediction of Unobserved Voxels from a Single Depth Image (2016, background): “We store the medoid for efficiency reasons, but it is also possible to store multiple modes, e.g. [18] .”
Latent Regression Forest: Structured Estimation of 3D Articulated Hand Posture (2014, background): “Many works for human body pose estimation use large synthetic datasets in training and use either pixel-wise classification [15], or joint regression techniques [5, 18, 13, 16] for pose estimation.”
Metric Regression Forests for Correspondence Estimation (2015, methodology): “Motivated by the90 success of Hough forests (Gall et al. 2011) for object detec-91 tion and localization, a follow-up paper (Girshick et al. 2011)92 directly regressed at each pixel an offset to several joint loca-93 tions.”
Latent-Class Hough Forests for 3D Object Detection and Pose Estimation (2014, methodology): “Following the approach of Girshick et al.[29] we only store the modes of the distribution which we find efficiently via the mean shift algorithm.”
Learning 6D Object Pose Estimation Using 3D Object Coordinates (2014, methodology): “The quantization allows us to use the standard information gain classification objective during training, which has the ability to cope better with the often heavily multi-model distributions p(y|c) than a regression objective [6].”